fs/iomap.c

   1 /*
   2  * Copyright (C) 2010 Red Hat, Inc.
   3  * Copyright (c) 2016 Christoph Hellwig.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14 #include <linux/module.h>
  15 #include <linux/compiler.h>
  16 #include <linux/fs.h>
  17 #include <linux/iomap.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/gfp.h>
  20 #include <linux/mm.h>
  21 #include <linux/swap.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/file.h>
  24 #include <linux/uio.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/buffer_head.h>
  27 #include <linux/dax.h>
  28 #include "internal.h"
  29
  30 typedef loff_t (*iomap_actor_t)(struct inode *inode, loff_t pos, loff_t len,
  31                 void *data, struct iomap *iomap);
  32
  33 /*
  34  * Execute a iomap write on a segment of the mapping that spans a
  35  * contiguous range of pages that have identical block mapping state.
  36  *
  37  * This avoids the need to map pages individually, do individual allocations
  38  * for each page and most importantly avoid the need for filesystem specific
  39  * locking per page. Instead, all the operations are amortised over the entire
  40  * range of pages. It is assumed that the filesystems will lock whatever
  41  * resources they require in the iomap_begin call, and release them in the
  42  * iomap_end call.
  43  */
  44 static loff_t
  45 iomap_apply(struct inode *inode, loff_t pos, loff_t length, unsigned flags,
  46                 struct iomap_ops *ops, void *data, iomap_actor_t actor)
  47 {
  48         struct iomap iomap = { 0 };
  49         loff_t written = 0, ret;
  50
  51         /*
  52          * Need to map a range from start position for length bytes. This can
  53          * span multiple pages - it is only guaranteed to return a range of a
  54          * single type of pages (e.g. all into a hole, all mapped or all
  55          * unwritten). Failure at this point has nothing to undo.
  56          *
  57          * If allocation is required for this range, reserve the space now so
  58          * that the allocation is guaranteed to succeed later on. Once we copy
  59          * the data into the page cache pages, then we cannot fail otherwise we
  60          * expose transient stale data. If the reserve fails, we can safely
  61          * back out at this point as there is nothing to undo.
  62          */
  63         ret = ops->iomap_begin(inode, pos, length, flags, &iomap);
  64         if (ret)
  65                 return ret;
  66         if (WARN_ON(iomap.offset > pos))
  67                 return -EIO;
  68
  69         /*
  70          * Cut down the length to the one actually provided by the filesystem,
  71          * as it might not be able to give us the whole size that we requested.
  72          */
  73         if (iomap.offset + iomap.length < pos + length)
  74                 length = iomap.offset + iomap.length - pos;
  75
  76         /*
  77          * Now that we have guaranteed that the space allocation will succeed.
  78          * we can do the copy-in page by page without having to worry about
  79          * failures exposing transient data.
  80          */
  81         written = actor(inode, pos, length, data, &iomap);
  82
  83         /*
  84          * Now the data has been copied, commit the range we've copied.  This
  85          * should not fail unless the filesystem has had a fatal error.
  86          */
  87         ret = ops->iomap_end(inode, pos, length, written > 0 ? written : 0,
  88                         flags, &iomap);
  89
  90         return written ? written : ret;
  91 }
  92
  93 static void
  94 iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
  95 {
  96         loff_t i_size = i_size_read(inode);
  97
  98         /*
  99          * Only truncate newly allocated pages beyoned EOF, even if the
 100          * write started inside the existing inode size.
 101          */
 102         if (pos + len > i_size)
 103                 truncate_pagecache_range(inode, max(pos, i_size), pos + len);
 104 }
 105
 106 static int
 107 iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags,
 108                 struct page **pagep, struct iomap *iomap)
 109 {
 110         pgoff_t index = pos >> PAGE_SHIFT;
 111         struct page *page;
 112         int status = 0;
 113
 114         BUG_ON(pos + len > iomap->offset + iomap->length);
 115
 116         page = grab_cache_page_write_begin(inode->i_mapping, index, flags);
 117         if (!page)
 118                 return -ENOMEM;
 119
 120         status = __block_write_begin_int(page, pos, len, NULL, iomap);
 121         if (unlikely(status)) {
 122                 unlock_page(page);
 123                 put_page(page);
 124                 page = NULL;
 125
 126                 iomap_write_failed(inode, pos, len);
 127         }
 128
 129         *pagep = page;
 130         return status;
 131 }
 132
 133 static int
 134 iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
 135                 unsigned copied, struct page *page)
 136 {
 137         int ret;
 138
 139         ret = generic_write_end(NULL, inode->i_mapping, pos, len,
 140                         copied, page, NULL);
 141         if (ret < len)
 142                 iomap_write_failed(inode, pos, len);
 143         return ret;
 144 }
 145
 146 static loff_t
 147 iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 148                 struct iomap *iomap)
 149 {
 150         struct iov_iter *i = data;
 151         long status = 0;
 152         ssize_t written = 0;
 153         unsigned int flags = AOP_FLAG_NOFS;
 154
 155         /*
 156          * Copies from kernel address space cannot fail (NFSD is a big user).
 157          */
 158         if (!iter_is_iovec(i))
 159                 flags |= AOP_FLAG_UNINTERRUPTIBLE;
 160
 161         do {
 162                 struct page *page;
 163                 unsigned long offset;   /* Offset into pagecache page */
 164                 unsigned long bytes;    /* Bytes to write to page */
 165                 size_t copied;          /* Bytes copied from user */
 166
 167                 offset = (pos & (PAGE_SIZE - 1));
 168                 bytes = min_t(unsigned long, PAGE_SIZE - offset,
 169                                                 iov_iter_count(i));
 170 again:
 171                 if (bytes > length)
 172                         bytes = length;
 173
 174                 /*
 175                  * Bring in the user page that we will copy from _first_.
 176                  * Otherwise there's a nasty deadlock on copying from the
 177                  * same page as we're writing to, without it being marked
 178                  * up-to-date.
 179                  *
 180                  * Not only is this an optimisation, but it is also required
 181                  * to check that the address is actually valid, when atomic
 182                  * usercopies are used, below.
 183                  */
 184                 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
 185                         status = -EFAULT;
 186                         break;
 187                 }
 188
 189                 status = iomap_write_begin(inode, pos, bytes, flags, &page,
 190                                 iomap);
 191                 if (unlikely(status))
 192                         break;
 193
 194                 if (mapping_writably_mapped(inode->i_mapping))
 195                         flush_dcache_page(page);
 196
 197                 pagefault_disable();
 198                 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
 199                 pagefault_enable();
 200
 201                 flush_dcache_page(page);
 202                 mark_page_accessed(page);
 203
 204                 status = iomap_write_end(inode, pos, bytes, copied, page);
 205                 if (unlikely(status < 0))
 206                         break;
 207                 copied = status;
 208
 209                 cond_resched();
 210
 211                 iov_iter_advance(i, copied);
 212                 if (unlikely(copied == 0)) {
 213                         /*
 214                          * If we were unable to copy any data at all, we must
 215                          * fall back to a single segment length write.
 216                          *
 217                          * If we didn't fallback here, we could livelock
 218                          * because not all segments in the iov can be copied at
 219                          * once without a pagefault.
 220                          */
 221                         bytes = min_t(unsigned long, PAGE_SIZE - offset,
 222                                                 iov_iter_single_seg_count(i));
 223                         goto again;
 224                 }
 225                 pos += copied;
 226                 written += copied;
 227                 length -= copied;
 228
 229                 balance_dirty_pages_ratelimited(inode->i_mapping);
 230         } while (iov_iter_count(i) && length);
 231
 232         return written ? written : status;
 233 }
 234
 235 ssize_t
 236 iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter,
 237                 struct iomap_ops *ops)
 238 {
 239         struct inode *inode = iocb->ki_filp->f_mapping->host;
 240         loff_t pos = iocb->ki_pos, ret = 0, written = 0;
 241
 242         while (iov_iter_count(iter)) {
 243                 ret = iomap_apply(inode, pos, iov_iter_count(iter),
 244                                 IOMAP_WRITE, ops, iter, iomap_write_actor);
 245                 if (ret <= 0)
 246                         break;
 247                 pos += ret;
 248                 written += ret;
 249         }
 250
 251         return written ? written : ret;
 252 }
 253 EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
 254
 255 static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset,
 256                 unsigned bytes, struct iomap *iomap)
 257 {
 258         struct page *page;
 259         int status;
 260
 261         status = iomap_write_begin(inode, pos, bytes,
 262                         AOP_FLAG_UNINTERRUPTIBLE | AOP_FLAG_NOFS, &page, iomap);
 263         if (status)
 264                 return status;
 265
 266         zero_user(page, offset, bytes);
 267         mark_page_accessed(page);
 268
 269         return iomap_write_end(inode, pos, bytes, bytes, page);
 270 }
 271
 272 static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes,
 273                 struct iomap *iomap)
 274 {
 275         sector_t sector = iomap->blkno +
 276                 (((pos & ~(PAGE_SIZE - 1)) - iomap->offset) >> 9);
 277
 278         return __dax_zero_page_range(iomap->bdev, sector, offset, bytes);
 279 }
 280
 281 static loff_t
 282 iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count,
 283                 void *data, struct iomap *iomap)
 284 {
 285         bool *did_zero = data;
 286         loff_t written = 0;
 287         int status;
 288
 289         /* already zeroed?  we're done. */
 290         if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN)
 291                 return count;
 292
 293         do {
 294                 unsigned offset, bytes;
 295
 296                 offset = pos & (PAGE_SIZE - 1); /* Within page */
 297                 bytes = min_t(unsigned, PAGE_SIZE - offset, count);
 298
 299                 if (IS_DAX(inode))
 300                         status = iomap_dax_zero(pos, offset, bytes, iomap);
 301                 else
 302                         status = iomap_zero(inode, pos, offset, bytes, iomap);
 303                 if (status < 0)
 304                         return status;
 305
 306                 pos += bytes;
 307                 count -= bytes;
 308                 written += bytes;
 309                 if (did_zero)
 310                         *did_zero = true;
 311         } while (count > 0);
 312
 313         return written;
 314 }
 315
 316 int
 317 iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
 318                 struct iomap_ops *ops)
 319 {
 320         loff_t ret;
 321
 322         while (len > 0) {
 323                 ret = iomap_apply(inode, pos, len, IOMAP_ZERO,
 324                                 ops, did_zero, iomap_zero_range_actor);
 325                 if (ret <= 0)
 326                         return ret;
 327
 328                 pos += ret;
 329                 len -= ret;
 330         }
 331
 332         return 0;
 333 }
 334 EXPORT_SYMBOL_GPL(iomap_zero_range);
 335
 336 int
 337 iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
 338                 struct iomap_ops *ops)
 339 {
 340         unsigned blocksize = (1 << inode->i_blkbits);
 341         unsigned off = pos & (blocksize - 1);
 342
 343         /* Block boundary? Nothing to do */
 344         if (!off)
 345                 return 0;
 346         return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
 347 }
 348 EXPORT_SYMBOL_GPL(iomap_truncate_page);
 349
 350 static loff_t
 351 iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
 352                 void *data, struct iomap *iomap)
 353 {
 354         struct page *page = data;
 355         int ret;
 356
 357         ret = __block_write_begin_int(page, pos & ~PAGE_MASK, length,
 358                         NULL, iomap);
 359         if (ret)
 360                 return ret;
 361
 362         block_commit_write(page, 0, length);
 363         return length;
 364 }
 365
 366 int iomap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 367                 struct iomap_ops *ops)
 368 {
 369         struct page *page = vmf->page;
 370         struct inode *inode = file_inode(vma->vm_file);
 371         unsigned long length;
 372         loff_t offset, size;
 373         ssize_t ret;
 374
 375         lock_page(page);
 376         size = i_size_read(inode);
 377         if ((page->mapping != inode->i_mapping) ||
 378             (page_offset(page) > size)) {
 379                 /* We overload EFAULT to mean page got truncated */
 380                 ret = -EFAULT;
 381                 goto out_unlock;
 382         }
 383
 384         /* page is wholly or partially inside EOF */
 385         if (((page->index + 1) << PAGE_SHIFT) > size)
 386                 length = size & ~PAGE_MASK;
 387         else
 388                 length = PAGE_SIZE;
 389
 390         offset = page_offset(page);
 391         while (length > 0) {
 392                 ret = iomap_apply(inode, offset, length, IOMAP_WRITE,
 393                                 ops, page, iomap_page_mkwrite_actor);
 394                 if (unlikely(ret <= 0))
 395                         goto out_unlock;
 396                 offset += ret;
 397                 length -= ret;
 398         }
 399
 400         set_page_dirty(page);
 401         wait_for_stable_page(page);
 402         return 0;
 403 out_unlock:
 404         unlock_page(page);
 405         return ret;
 406 }
 407 EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
 408
 409 struct fiemap_ctx {
 410         struct fiemap_extent_info *fi;
 411         struct iomap prev;
 412 };
 413
 414 static int iomap_to_fiemap(struct fiemap_extent_info *fi,
 415                 struct iomap *iomap, u32 flags)
 416 {
 417         switch (iomap->type) {
 418         case IOMAP_HOLE:
 419                 /* skip holes */
 420                 return 0;
 421         case IOMAP_DELALLOC:
 422                 flags |= FIEMAP_EXTENT_DELALLOC | FIEMAP_EXTENT_UNKNOWN;
 423                 break;
 424         case IOMAP_UNWRITTEN:
 425                 flags |= FIEMAP_EXTENT_UNWRITTEN;
 426                 break;
 427         case IOMAP_MAPPED:
 428                 break;
 429         }
 430
 431         return fiemap_fill_next_extent(fi, iomap->offset,
 432                         iomap->blkno != IOMAP_NULL_BLOCK ? iomap->blkno << 9: 0,
 433                         iomap->length, flags | FIEMAP_EXTENT_MERGED);
 434
 435 }
 436
 437 static loff_t
 438 iomap_fiemap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 439                 struct iomap *iomap)
 440 {
 441         struct fiemap_ctx *ctx = data;
 442         loff_t ret = length;
 443
 444         if (iomap->type == IOMAP_HOLE)
 445                 return length;
 446
 447         ret = iomap_to_fiemap(ctx->fi, &ctx->prev, 0);
 448         ctx->prev = *iomap;
 449         switch (ret) {
 450         case 0:         /* success */
 451                 return length;
 452         case 1:         /* extent array full */
 453                 return 0;
 454         default:
 455                 return ret;
 456         }
 457 }
 458
 459 int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fi,
 460                 loff_t start, loff_t len, struct iomap_ops *ops)
 461 {
 462         struct fiemap_ctx ctx;
 463         loff_t ret;
 464
 465         memset(&ctx, 0, sizeof(ctx));
 466         ctx.fi = fi;
 467         ctx.prev.type = IOMAP_HOLE;
 468
 469         ret = fiemap_check_flags(fi, FIEMAP_FLAG_SYNC);
 470         if (ret)
 471                 return ret;
 472
 473         ret = filemap_write_and_wait(inode->i_mapping);
 474         if (ret)
 475                 return ret;
 476
 477         while (len > 0) {
 478                 ret = iomap_apply(inode, start, len, 0, ops, &ctx,
 479                                 iomap_fiemap_actor);
 480                 if (ret < 0)
 481                         return ret;
 482                 if (ret == 0)
 483                         break;
 484
 485                 start += ret;
 486                 len -= ret;
 487         }
 488
 489         if (ctx.prev.type != IOMAP_HOLE) {
 490                 ret = iomap_to_fiemap(fi, &ctx.prev, FIEMAP_EXTENT_LAST);
 491                 if (ret < 0)
 492                         return ret;
 493         }
 494
 495         return 0;
 496 }
 497 EXPORT_SYMBOL_GPL(iomap_fiemap);